China's top court sets the rules of the road for AI deepfake lawsuits
China's Supreme People's Court put its judicial weight behind the "AI face-swap / voice-clone" problem on Monday, releasing a 24-article opinion that tells lower courts exactly how to rule on deepfake suits, AI hallucination claims, "doxxing-by-AI" cases, algorithmic price discrimination, autonomous-driving liability, model-training disputes, and open-source software. The opinion is the SPC's broadest attempt yet to translate the country's existing civil code into a working playbook for synthetic-media harm — and it lands as telecom fraud built on cloned faces and voices is becoming, in the court's own words, "more intelligent, more structured, and more socially harmful."
What the opinion actually does: it gives judges a unified framework for assigning liability when a deepfake injures a person's portrait, voice, or reputation; lays out evidentiary rules for proving that a synthetic output is what caused the harm; addresses the open question of who pays when a model is trained on data without consent; and folds in a new "AI hallucination" tort, so courts have somewhere to slot the growing class of "the model said something defamatory about me" claims. SPC Third Criminal Tribunal head Wang Bin used the press conference to call out the "refined black-grey industry chain" — organized, industrial-scale fraud shops that have built their workflows around off-the-shelf face-swap and voice-clone models. The data point he leaned on: handled telecom-fraud cases increased again in 2025.
Why it matters outside China: the SPC opinion is the most concrete national-court answer yet to the question every jurisdiction is now wrestling with — when a generative model produces harm, who is on the hook, and on what timeline? China is choosing to answer that through the civil bench rather than through new legislation, which is faster but leaves more to judge-by-judge interpretation. U.S. and EU courts have been slow to commit to analogous defaults; the SPC's move puts pressure on platforms and model vendors to disclose their training data, because the opinion treats training without lawful basis as a potential infringement. Expect Chinese courts to start citing this opinion within weeks, and expect foreign platforms operating in China to start getting sued in it.
Mprosevir becomes China's first AI-assisted drug to reach the market. Westlake University, Westlake Laboratory, and Westlake Pharmaceuticals won conditional NMPA approval for the small-molecule COVID-19 therapy, a Class-1 designation reserved for genuinely new chemical entities with novel mechanisms. The discovery workflow combined DNA-encoded library screening of 49 billion compounds with an AI virtual-screening pass that narrowed more than 100 candidates down to nine in days; six of those showed real activity, and the team picked WLU6937 as the lead. The drug is the first approved molecule to come out of the 34-year-old DEL technology, and the first to market where AI was load-bearing across both the screening and the optimization stages. The bigger signal: China's regulator is willing to greenlight a drug whose primary discovery engine was a model, not a bench — a different category of green light than a lab demo, a benchmark, or a clinical-trial enrollment.
What to watch: whether the Mprosevir approval triggers follow-on NMPA decisions for the half-dozen AI-discovered candidates now in Chinese trials, and whether the SPC's deepfake opinion gets cited by a court in a non-Chinese case as a comparative-law reference.
Should AI model vendors carry strict liability for outputs that cause real-world harm, or is a negligence standard still the right default? Tell us in the comments.
